This study presents the design and implementation of a quadruped robot platform supporting Sim-to-Real reinforcement learning (RL) applications. The system encompasses electromechanical hardware design, kinematic modeling, simulation environment construction, and RL-based control training. The platform is intended as a locally developed solution for academic research in Taiwan, offering an affordable and open alternative to foreign commercial robots with limited access to low-level controls.

The proposed locomotion of Quasar quadruped robot is based on RL techniques. The RL models are trained with simulation using NVIDIA Isaac Sim and Isaac Lab. Final experiments confirm that the trained models enable stable walking on flat terrain, with analysis of the performance gap between simulation and real-world results.

The quadruped robot has high potential to be applicable for industrial inspection tasks to resolve the labor storage problems.